The Role of the Range Parameter for Estimation and Prediction in Geostatistics

TL;DR

This paper demonstrates that jointly estimating the range and variance parameters in geostatistics improves finite-sample performance and the applicability of asymptotic results, especially for processes with small effective ranges.

Contribution

It shows that joint estimation of range and variance yields better finite-sample results and more accurate asymptotic approximations than fixing the range parameter.

Findings

Joint estimation improves prediction accuracy.

Asymptotic results are more applicable with joint estimation.

Performance gains are notable for small effective ranges.

Abstract

Two canonical problems in geostatistics are estimating the parameters in a specified family of stochastic process models and predicting the process at new locations. A number of asymptotic results addressing these problems over a fixed spatial domain indicate that, for a Gaussian process with Mat\'ern covariance function, one can fix the range parameter controlling the rate of decay of the process and obtain results that are asymptotically equivalent to the case that the range parameter is known. In this paper we show that the same asymptotic results can be obtained by jointly estimating both the range and the variance of the process using maximum likelihood or maximum tapered likelihood. Moreover, we show that intuition and approximations derived from asymptotic arguments using a fixed range parameter can be problematic when applied to finite samples, even for moderate to large sample sizes. In contrast, we show via simulation that performance on a variety of metrics is improved and asymptotic approximations are applicable for smaller sample sizes when the range and variance parameters are jointly estimated. These effects are particularly apparent when the process is mean square differentiable or the effective range of spatial correlation is small.